How to Write Chatbot Use Cases (With Examples and a Template)

A chatbot use case is one specific job your chatbot does for one kind of visitor: who is asking, what they want, what the bot answers from, what a good reply looks like, and when a person takes over. Writing chatbot use cases means filling in those parts for each situation before you set the bot up, so you know what it should say, what content it needs, and how you'll tell whether it works.

I build ChatBento, an AI support chatbot for websites. A one-page list of use cases takes about an hour, and it tells you which pages to fix, which instructions to write and which questions to test. Here's the template I use, six filled-in examples, and how to find, rank and test your own.

What makes a good chatbot use case?

A use case is narrow. "Customer support" is not a use case. "A shopper who bought something in the sale asks whether they can return it" is. A good one has a real visitor you've actually heard from, an answer written down somewhere you control, and a clear end: an answer or a person. If you couldn't answer it by email from your own site, the bot can't either.

The chatbot use case template

Copy this block once per use case. Keep each line to a sentence or two.

Use case:       [short name, e.g. "Returns on sale items"]
Visitor:        [who they are and which page they're on]
Goal / trigger: [what they want, and what made them open the chat]
Source content: [the pages, Q&As or pasted text that hold the answer]
Good answer:    [what the reply must include, and the link it should give]
Hand off when:  [when a person takes over, and what the visitor should include]
Measure:        [how you'll know it works, and where you'll look]

The trigger is what happened just before the chat: a price they didn't understand, a late order, a Sunday evening with nobody answering the phone. For source content, name the exact page or Q&A; if you can't, you've found a gap to fill before launch. The good answer line becomes your test. For handoff, say what the visitor should include so you can reply without asking again; my nine handoff rules cover the common triggers. And make the measure something you can count weekly, like "returns questions that got the right policy", not "customer satisfaction".

Six chatbot use case examples

Each example is a filled-in template plus a sample exchange. The businesses are made up, and every fact in the replies is assumed to be on that business's own site. The use cases hub covers twelve of these jobs in more depth.

1. Pre-sales question: a cleaning company

• Visitor: A homeowner on the pricing page, comparing a few cleaners.
• Goal / trigger: Wants to know whether you cover their area and what a deep clean costs.
• Source content: Pricing page, service-area page, a Q&A on what a deep clean includes.
• Good answer: The service area as written, the price, what's included, a link to the pricing page.
• Hand off when: They want a quote for a large job, or ask for a discount.
• Measure: Pricing questions in the inbox that got the right price and link.

Visitor: do you do deep cleans in oakfield and how much

Bot: Yes, Oakfield is in our service area. A deep clean starts at $180 for up to two bedrooms and includes the inside of the oven and fridge. The full price list is on our pricing page. For a bigger home or a specific date, I can pass your details to the team.

The pre-sales questions use case shows how to set this one up.

2. After-hours support: an auto repair shop

• Visitor: A driver on the homepage at 9 p.m., after a warning light came on.
• Goal / trigger: Wants to know whether the shop is open on Saturday and whether to book.
• Source content: Hours on the contact page, a Q&A on walk-ins, a Q&A on reply times.
• Good answer: Saturday hours, the walk-in rule, the booking link, when the team replies.
• Hand off when: They want a price for a specific repair or a call back.
• Measure: Out-of-hours conversations that ended with a correct answer, and overnight leads.

Visitor: are you open saturday? check engine light just came on

Bot: We're open Saturdays from 8 a.m. to 1 p.m. We take walk-ins for diagnostics, but booking online gets you a set time. I can't tell what's causing the light. If you'd like the team to get back to you, leave your email below and they'll reply on the next working day.

3. Lead capture: a wedding photographer

• Visitor: A couple on the packages page, planning a wedding a year away.
• Goal / trigger: Want prices and to know whether their date is free.
• Source content: Packages page, a Q&A on travel fees, a Q&A on reserving a date.
• Good answer: Packages and prices, and an honest line that the bot can't see the calendar.
• Hand off when: They ask about a specific date. That's the lead: their email, plus the date and venue in the note.
• Measure: Leads per month, and how many became bookings.

Visitor: how much for a full day and are you free 12 june 2027?

Bot: The full-day package is $2,400 and covers preparations through to the first dance, with an online gallery. I can't see the calendar, so I can't confirm 12 June. Leave your email below with your venue, and the team will check the date and reply.

4. Returns policy: an online clothing store

• Visitor: A customer on the returns page, a week after buying.
• Goal / trigger: A sale item doesn't fit, and they want to know if it can go back.
• Source content: Returns page, plus a Q&A that states the sale-item rule word for word.
• Good answer: The time limit, the sale-item rule as written, a link to the returns page.
• Hand off when: The item is faulty, they want an exception, or their order needs checking. Ask for the order number.
• Measure: Returns emails per week, before and after launch.

Visitor: can i return something i bought in the sale, its too small

Bot: Sale items can be exchanged for a different size within 30 days, but they can't be refunded. The steps are on our returns page. If the item is faulty, or you'd like the team to check your order, leave your email and order number below.

5. Booking questions: a hair salon

• Visitor: Someone on the services page deciding whether to book.
• Goal / trigger: Wants to know how long a color appointment takes, what it costs, and whether Saturday is free.
• Source content: Services and prices page, a Q&A on deposits and cancellations, a "How do I book?" Q&A with the booking link.
• Good answer: Duration, price, the deposit rule, and the booking page for live availability.
• Hand off when: They mention a reaction to a past treatment, or want to change a booking.
• Measure: Routine booking calls, tallied for a week before and after.

Visitor: how long is balayage and can i get in saturday

Bot: Balayage takes about three hours and starts at $165, depending on hair length. I can't see the diary, but our booking page shows live availability for Saturday. There's a $30 deposit, refunded if you cancel at least 48 hours before.

The bot answers; your booking page or a person books. The booking questions use case explains why.

6. SaaS how-to: an invoicing app

• Visitor: A trial user in their first week.
• Goal / trigger: Wants their logo on invoices and can't find the setting.
• Source content: The help article on invoice branding, plus pasted setup notes that aren't published.
• Good answer: The steps in short form and a link to the full article.
• Hand off when: Their account doesn't behave as the docs describe, or it's about billing.
• Measure: How-to tickets from trial users.

Visitor: how do i put my logo on invoices

Bot: Go to Settings, then Branding, and upload your logo under Invoice appearance. It appears on new invoices; ones you've already sent keep the old design. The invoice branding guide has screenshots of each step.

How do you find your use cases?

Write down real questions, in the customer's own words, from:

• Past emails. The ones you answered with copy and paste are your best use cases.
• Contact-form submissions. Each is a visitor who couldn't find an answer on the page they were on.
• Chat transcripts, if you've used a chat tool before.
• Search Console queries. Google's Performance report groups your data by the search query users typed. Queries starting "do you", "how much" or "can I" are questions people have before they reach your site.
• Sales and phone calls. Ask whoever answers the phone for the five questions they hear every week.

Group the questions that share an answer. If you have questions but no written answers, the free FAQ generator gives you a first draft to edit into Q&As.

How do you prioritize them?

Score each use case from 1 to 3 on three things, then multiply.

Start with the highest scores. The cleaning company's pricing question might score 3 × 3 × 2 = 18; a refund on a damaged order, 2 × 1 × 1 = 2.

A low score doesn't mean you drop it. A low "answerable" score means writing the answer down first. A 1 for cost of a wrong answer makes it a handoff use case: the bot explains the general policy and gets the visitor to a person.

Common mistakes

Use cases the bot can't fulfil. "The bot books the appointment" or "checks where my order is". A bot that answers from your website isn't connected to your booking or order systems. Rewrite these as "the bot answers, then a person or your booking page acts", or you get a bot describing an action that never happens.

Vague goals. "Improve the customer experience" can't be tested. "A shopper gets the sale-item return rule" can.

No handoff rule. Without an end, the bot keeps trying to answer something it can't, and the visitor goes round in circles. A clear human handoff point belongs in every template.

No source content. The most common one: "the bot explains our cancellation policy", when the policy only exists in the owner's head. Write it down as a page or Q&A first. Instructions shape how the bot answers, not what it knows; my guide to writing chatbot instructions covers what goes where.

How do you test each use case before launch?

Your "Good answer" and "Hand off when" lines are the test. For each use case, ask the bot three questions:

1. The question as a customer actually wrote it.
2. The same question in other words, with a typo.
3. A version that should trigger the handoff.

The first two pass when the reply contains everything on the "Good answer" line; the third when the bot offers the contact form instead of guessing. When something fails, the fix is usually content: a missing Q&A or an outdated page. Then run my 30-question launch test for what your use cases don't cover, such as off-topic requests and tricks.

After launch, the "Measure" line takes over. In ChatBento, the inbox holds every conversation and the unanswered questions list shows what the bot couldn't answer. Read both weekly for the first month. Resolution, not deflection explains which numbers are worth counting.

Where ChatBento fits

ChatBento covers the use cases answered from your content, like the six above. It trains on your pages, Q&As and pasted text, installs with one script tag, and answers around the clock from that content only. When it doesn't know, it says so and offers a contact form (email required, name and note optional) that's emailed to you. It doesn't take actions: no booking, order lookup, live chat, ticketing, WhatsApp or CRM integrations. If your top use cases need those, it's the wrong tool. If they're answers and handoffs, the free plan has 50 AI messages a month, enough to test each of your top use cases, and Hobby is $14 a month.